Industrial communications
Replacing hundreds of point-to-point cables with a twisted pair that runs through the plant is not just a saving in copper: it is what allows each device to report its status, diagnostics to be done remotely and production to be measured.
The physical media—RS-485, CAN, differential pair, termination, isolation—are covered in serial communication buses. Here we work on what goes on top: industrial protocols and the architecture of a plant network.
01The automation pyramid
| Level | What is there | Required time |
|---|---|---|
| Field | Sensors, actuators, drives, remote I/O modules. | 1 a 10 ms |
| Control | PLCs and controllers that run the logic of each machine. | 10 a 100 ms |
| Supervision | SCADA, operator screens, historical logging and alarms. | 0.1 to 1 s |
| Management | Production, maintenance, quality, costs. | minutes or hours |
What distinguishes an industrial network from an office one is not that it is faster: it is that the delivery time is bounded and guaranteed. A message that sometimes takes 2 ms and sometimes 150 ms is useless for control, even if on average it looks extremely fast. That is why fieldbuses use orderly access mechanisms—cyclic polling, priorities, time slots—instead of letting each node transmit whenever it wants.
02Fieldbuses
| Bus | Medium | Nodes | Characteristics |
|---|---|---|---|
| Modbus RTU | RS-485 | 32 per segment | Extremely simple, open and universal. A master polls and the slaves respond. Any device can implement it. |
| Profibus DP | RS-485 | 126 | Very widespread in Europe. Deterministic cyclic exchange with profiles for each device type. |
| CANopen | CAN | 127 | On top of CAN: priorities and error detection built into the bus itself, with a standardized object profile. |
| DeviceNet | CAN | 64 | Also on top of CAN, with node power on the same cable. |
| AS-i | 2 wires | 31 or 62 | For the lowest level: binary sensors and actuators. Data and power over the same flat cable, up to 100 m. |
| IO-Link | 3 wires | point to point | Communicates with each sensor over the usual cable: besides the value, it reports status, parameters and diagnostics. |
A yellow flat cable, with a profile that prevents reversing the polarity, to which the modules connect by piercing it with needle contacts: there is no need to strip or cut. Over those two wires travel both the data and the power for the sensors. A master polls the 31 slaves in a 5 ms cycle, guaranteed.
The calculation that justifies it: connecting 31 sensors point to point to a panel 40 m away takes about 2.5 km of cable and 62 terminals. With AS-i it is 40 m of flat cable and no terminals. That is the economic argument for every fieldbus, and it is seen more clearly here than anywhere else.
03Industrial Ethernet
| Protocol | How it achieves determinism |
|---|---|
| Modbus TCP | Modbus inside standard Ethernet. Simple and universal, but with no timing guarantees: suitable for supervision, not for fast control. |
| Profinet | Real-time channels that bypass part of the protocol stack, with cycles of a few milliseconds. |
| EtherNet/IP | Ethernet with the CIP protocol on top, widely used in the American market. |
| EtherCAT | Each node reads and writes the frame on the fly, as it passes through. It achieves cycles below one millisecond with dozens of nodes. |
- M12 connectors with IP67 rating instead of bare RJ45.
- Industrial DIN-rail switches, powered at 24 V and fanless.
- Cable with an oil-resistant jacket, rated for continuous flexing.
- Ring with fast recovery: on a break, the network reconfigures in milliseconds.
- Segmentation: the control network separated from the administrative network, with a firewall between the two.
A PLC reachable from the internet is a serious problem: most have no real authentication and allow anyone to stop the process or change the program.
The minimum measures: an isolated control network, remote access only through an encrypted tunnel, factory-default passwords changed, unused ports disabled and a log of who connects. And a backup of each device’s program, stored off-site.
04Commissioning and diagnostics
| Symptom | What to check |
|---|---|
| A node does not respond | Duplicate or misconfigured address, a speed different from the rest, or the node’s power supply. |
| Sporadic errors that grow with speed | Termination: 120 Ω at both ends and only there. |
| Faults when a motor starts | The bus cable runs alongside power cables, or galvanic isolation is missing. |
| The network works but control degrades | Network cycle time too long for what the process needs. Check how much data is exchanged and how often. |
| Everything fails since a device was added | Stub too long, or a segment that exceeded the allowed number of nodes. |
| Intermittent faults in motion | Ordinary cable in a cable carrier chain: cable rated for continuous flexing is needed. |
A drawing of the real topology, a node table with address, name, physical location and firmware version, lengths of each section and location of the terminations. Without that table, adding a device two years later means guessing free addresses with the line stopped.
05In the lab
Configure one PLC as master and another as slave over Modbus RTU. Exchange bits and words, and capture the frame with an analyzer or a PC converter to identify address, function, data and redundancy check (CRC).
With an output that toggles every time new data arrives from the slave, measure the actual update time with the oscilloscope. Add nodes and more data, and see how it grows. Compare with what the process would need.
On an assembled network: remove the termination, duplicate an address, set one node to a different speed and run the cable next to a running motor. Record the symptom in each case and build your own diagnostic table.
Connect a simple SCADA or an embedded web server that shows PLC variables in real time, with alarms and historical logging. Discuss what is published upward and what must stay in the control network, and why.
06Common mistakes
| Mistake | Consequence |
|---|---|
| Duplicate addresses | Both nodes respond at once and the bus becomes erratic. |
| Termination on every node or on none | Reflections and errors that worsen with speed and length. |
| Bus next to power cables | Faults that appear exactly when a motor starts. |
| Using Modbus TCP for fast control | It does not guarantee timing: suitable for supervision, not for control loops. |
| Control network joined to the administrative one | An office IT problem stops production, and the PLC is exposed. |
| PLC reachable from the internet | Anyone can stop the process or change the program. |
| Rigid cable in a cable carrier chain | It breaks from fatigue and causes intermittent faults that are very hard to locate. |
| No node table or backups | Every expansion or replacement is done blind and with the line stopped. |
07Self-assessment
What are the levels of the automation pyramid?
Field, control, supervision and management. The lower down, the tighter the timing requirement; the higher up, the more data volume and the less urgency.
What does it mean for a network to be deterministic?
That the delivery time is bounded and guaranteed. The average does not matter: the worst case does.
What makes the AS-i bus special?
It carries data and power over a two-wire flat cable, to which the modules connect by piercing it. It polls 31 slaves in a guaranteed 5 ms cycle.
Why does a fieldbus save wiring?
Because it replaces the point-to-point cables of each sensor with a single shared cable: 31 sensors at 40 m go from about 2.5 km of cable to 40 m.
How does EtherCAT achieve cycles below one millisecond?
Each node reads and writes the frame on the fly, as it passes, instead of receiving it, processing it and forwarding it.
What advantage does the ring topology have?
It tolerates a break: the network reconfigures and keeps working. In a line, a break splits it in two.
Why is the control network separated from the administrative one?
So that an office IT problem does not stop production and so that the control devices, which have almost no protection of their own, are not exposed.
Is Modbus TCP suitable for a fast control loop?
No: since it runs over standard Ethernet it does not guarantee delivery times. It is suitable for supervision and for non-critical exchange.
A new node made the whole network fail. What is checked first?
Duplicate address, a speed different from the rest, stub length and whether the allowed number of nodes in the segment was exceeded.
What must the documentation of an industrial network contain?
Real topology, a node table with address, name, location and firmware version, length of each section and location of the terminations.